1. Field of the Invention
The present invention relates to a write back method for a RAID control system in a disk array subsystem.
2. Description of the Related Art
RAID (redundant array of independent/inexpensive disk drives) apparatuses are widely used as means for storing a very large amount of data handled by a computer system. A RAID 5 especially, is widely utilized for reasons such as a high cost performance ratio, a simple configuration accomplishing protection of data, et cetera.
Since the RAID 5 maintains a redundancy of data by using a parity disk, the data (i.e., parity data) therein needs to be updated when performing a write back processing.
Here, the write back processing means the processing of writing, to the disk, information no longer needed by a cache memory when writing new information into a cache memory within a RAID control apparatus of a disk.
FIG. 1 exemplifies data within a cache memory comprised by a RAID control apparatus in the case of updating all data stripes at the time of a write back processing.
An update data sent from a host computer, et cetera, to the RAID control apparatus is subjected to a striping (simply “striped” hereinafter) thereby. And, if data as the target of updating exists in a cache memory, the aforementioned data will be updated by the update data.
Here, one piece of data striped by a RAID control apparatus is called a “data stripe” in the following description.
The data 1 shown by FIG. 1 indicates the data of a RAID 5 comprising four disks (i.e., disks #0 through #3 shown by FIG. 1), with the data being striped across #0 through #3.
And the data 1 shown by FIG. 1 indicates the case of all the data stripes #0 through #2 being update data.
Since all the updating data stripes (i.e., data stripes #0 through #2) are in the cache memory, parity data (i.e., a data stripe #3) can be generated without performing a processing of reading from the disks #0 through #3. Then the generated data 1 is stored in the disks #0 through #3 (NB: the writing method is called a “Bandwidth Write method”).
FIG. 2 exemplifies data within a cache memory comprised by a RAID control apparatus in the case of updating a part of data stripes at the time of a write back processing.
Data 2 shown by FIG. 2, likewise FIG. 1, shows data of RAID 5 comprising four disks (i.e., disks #0 through #3 shown by FIG. 2), with the data being striped in #0 through #3.
And the data 2 shown by FIG. 2 indicates the case of only the data stripes #0 and #1 being update data.
In this case, since a part of data of the stripes is in the cache memory, it is necessary to generate parity data by reading pre-update data from the disks #0, #1 and #3 and obtaining an exclusive logical sum (simply called “XOR” hereinafter) with the update data (i.e., the data stripes #0 and #1), followed by storing the update data and the generated parity data in the disks #0, #1 and #3, respectively (this storing (i.e., writing) method is called a “Small Write method”).
As described above, at the time of a data update processing under the RAID 5, in particular updating a part of the update data (i.e., a write processing to a disk), it is necessary to read once the pre-update data and parity data from the disks and generate new parity data (called “update parity” hereinafter) corresponding to the update data, followed by writing the update parity as well as the update data to the disk.
That is, updating a part of data stripes has faced the problem of needing an extra process, such as reading a pre-update data from the disk, et cetera, as compared to the case of updating all the data stripes, thus resulting in slowing down the processing speed at the time of updating.
In the meantime, a laid-open Japanese patent application publication No. 07-311659 has disclosed a disk array apparatus for managing update data after a write processing by a host computer by storing the update data in a cache memory, and also managing a pre-updated data by using a cache management table if the pre-updated data is left in the cache memory in order to shorten the time required for a write back processing, thereby improving a throughput.